Adversarial Augmentation

Adversarial augmentation is a data augmentation technique that improves the robustness and generalization of machine learning models by generating training examples designed to be challenging for the model. Current research focuses on applying this technique across various domains, including image classification, video action recognition, and natural language processing, often incorporating attention mechanisms, Bayesian methods, or adversarial training frameworks within models like transformers and convolutional neural networks. This approach is significant because it enhances model performance on out-of-distribution data and improves resilience to adversarial attacks, leading to more reliable and robust systems in diverse real-world applications.

Papers

December 9, 2021

3D-VField: Adversarial Augmentation of Point Clouds for Domain Generalization in 3D Object Detection
Alexander Lehner, Stefano Gasperini, Alvaro Marcos-Ramiro, Michael Schmidt, Mohammad-Ali Nikouei Mahani, Nassir Navab, Benjamin Busam, Federico Tombari
Point Cloud 3D Object Detection Domain Generalization Object Shape 3D Object Detector Adversarial Augmentation 3D Field

November 24, 2021

Challenges of Adversarial Image Augmentations
Arno Blaas, Xavier Suau, Jason Ramapuram, Nicholas Apostoloff, Luca Zappella
Technical Challenge Adversarial AutoEncoder Adversarial Augmentation Augmentation Policy Optimal Adversarial Random Augmentation Random Data Augmentation

Adversarial Augmentation

Papers

3D-VField: Adversarial Augmentation of Point Clouds for Domain Generalization in 3D Object Detection

Challenges of Adversarial Image Augmentations